[00:15:01] <Reedy>	 message 	Count
[00:15:02] <Reedy>	 {method} {exception}	135
[00:15:02] <Reedy>	 Lowered set() TTL for {cachekey} due to replication lag.  	132
[00:15:06] <Reedy>	 That's unhelpful
[01:01:22] <tgr>	 is wfGetDB(1) the same as wfGetDB(DB_REPLICA)?
[01:01:31] <tgr>	 in practice, I mean
[01:02:21] <Reedy>	 No?
[01:02:22] <Reedy>	 	const DB_REPLICA = -1;
[01:02:51] <Reedy>	 dunno what it does with ohter numbers...
[01:03:09] <tgr>	 gets the DB with that connection index
[01:03:25] <tgr>	 no idea how those get assigned though
[01:03:55] <tgr>	 just found a fun bug where some code is passing an array, which PHP type juggles to 1
[06:41:57] <legoktm>	 TimStarling: re: https://bugs.php.net/bug.php?id=47540 was it fixed in 5.2? Can we safely remove the PHP_SAPI check in FileBackend::getScopedPHPBehaviorForOps() now?
[06:43:16] <legoktm>	 hm, that code is also copied to LBFactory::getScopedPHPBehaviorForCommit()...
[06:44:09] <legoktm>	 ...and LoadBalancer::getScopedPHPBehaviorForCommit()
[06:49:28] <legoktm>	 AaronSchulz: T184044
[06:49:28] <stashbot>	 T184044: getScopedPHPBehaviorForCommit is copied 3 times in MediaWiki - https://phabricator.wikimedia.org/T184044
[15:08:11] <RoanKattouw>	 tgr: IIRC it's either the order in which DB servers are defined in the config, or the order in which connections are opened
[15:08:33] <RoanKattouw>	 If it's the former it'll open a replica connection but always to the same replica
[15:27:06] <anomie>	 tgr|away: wfGetDB(1) is supposed to work as RoanKattouw says, which I think for LBFactoryMulti goes by the order in $wgLBFactoryConf['sectionLoads']. But there's a bug (introduced in https://gerrit.wikimedia.org/r/#/c/178747/) that makes anything that's not DB_MASTER be treated as DB_REPLICA unless you specifically specify invalid groups for the second parameter. https://gerrit.wikimedia.org/r/#/c/401520/ has a fix.
[15:29:54] <RoanKattouw>	 Ha nice catch
[15:30:11] <RoanKattouw>	 Only been broken for 3 years huh
[15:34:48] <DanielK_WMDE>	 Yea, I ran into that the other day when working on preventing writes to replicas... Anomie tracked down the origins of the issue.